# NOTCRM: Free Interactive Agentic AI Course for Product Managers

> An interactive simulation lab where Product Managers, AI Engineers, and Enterprise Leaders learn how multi-agent AI systems are architected, evaluated, governed, and operated in production.

- Website: https://notcrm.onrender.com/
- Target Audience: AI Product Managers, Technical PMs, Engineering Managers, AI Architects
- Prerequisites: None (Designed for non-technical and technical learners alike)
- Cost: 100% Free with downloadable Certificate of Completion
- Architecture Reference: Andrew Ng's 4 Agentic Workflow Patterns and Andrej Karpathy's LLM OS Principles

## Core Curriculum & Modules

### 1. Foundational Architecture
- Multi-Agent DAG Orchestration: Moving beyond monolithic prompts. Decomposing workflow tasks across specialized agents (Intake, Research, Qualification, Security, Commercial) with isolated error boundaries.
- Model Context Protocol (FastMCP): Decoupling reasoning models from external data sources and tools (CRM, Knowledge Base, Security APIs) using standardized, typed schemas.
- Deterministic Pipeline Execution: Coordinating sequential and parallel dependency graphs with human-in-the-loop (HITL) checkpoints.

### 2. FAANG Interview Readiness (Google, Meta, Amazon, Apple, Netflix)
- Agent-to-Agent (A2A) Protocols: Structured message passing and typed payload schemas that eliminate context leakage across agent transitions.
- Enterprise Knowledge Graph Memory: How systems retain episodic memory from historical outcomes, learning recurring risk factors without expensive continuous fine-tuning.
- Observability and Unit Economics: Tracking LiteLLM execution spans, p95 latency percentiles, token spend, and outcome-based pricing models (+credit for correct approvals, penalty for policy violations).

### 3. Frontier AI Labs Interview Readiness (Anthropic, OpenAI, DeepSeek)
- LLM Evaluations and Golden Datasets: Versioned benchmarks across failure taxonomies (clean, ambiguous, stale context, conflicting claims, adversarial inputs).
- Trajectory Scorecards: Scoring process quality, tool call efficiency, and penalizing unsafe execution paths.
- Deterministic Independent Verifiers: Rule-based claim-to-evidence verification running ahead of LLM consensus to prevent hallucinations and ungrounded commitments.
- Red-Teaming and Safety Guardrails: Input/output guardrails neutralizing prompt injections and guaranteeing SOC2, GDPR, and SOX compliance.

## Industry Foundations
- Andrew Ng (DeepLearning.AI): Implements the 4 foundational agentic design patterns: Reflection, Tool Use, Planning, and Multi-Agent Collaboration.
- Andrej Karpathy: Demonstrates the LLM OS paradigm: context window as RAM, LLM as CPU, FastMCP as I/O peripherals, and Knowledge Graphs as persistent disk storage.
- Anthropic: Aligned with the 'Building Effective Agents' framework (Workflows vs Autonomous Agents, Router, Orchestrator-Workers, Evaluator-Optimizer).
- Meta & Google: Follows AGENTS.md engineering principles for modular agent design, bounded memory, and deterministic verification.